Proj 13420 same variability as 13418

[HaloJones]

Me too now,

13420 (5363, 23, 0)
13420 (4786, 34, 0)

but also one fastish one

13420 (7724, 28, 0)

[Neil-B]

In my case standard length ones return 200% expected ppd .. I have had one or two slower ones giving 50-75% expected ppd ... overall for time spent folding all these project I am around 175% ppd.

@HaloJones from your posts 17 odd standard and a couple of slower ones ... what ppd are you getting for standard and slower compared to what you would expect from your kit?

[bruce]

John has a new staff member who has been assigned the responsibility for looking at all the performance reports for anomalies.

[Neil-B]

@bruce ... nice to know ... I'll wind my neck in !!

[HaloJones]

In my case standard length ones return 200% expected ppd .. I have had one or two slower ones giving 50-75% expected ppd ... overall for time spent folding all these project I am around 175% ppd.

@HaloJones from your posts 17 odd standard and a couple of slower ones ... what ppd are you getting for standard and slower compared to what you would expect from your kit?

According to HFM Benchmark Viewer:

Code: Select all

Name: i7-1070x2 Slot 00
 Path: 192.168.0.27:36330
 Proc: GPU / GeForce GTX 1070 (Linux Mint 18, NVidia drivers 415.27)
 Number of Frames Observed: 300

 Min. Time / Frame : 00:02:15 - 1,199,697 PPD
 Avg. Time / Frame : 00:02:35 - 975,156 PPD
 Cur. Time / Frame : 00:03:15 - 688,529 PPD
 R3F. Time / Frame : 00:03:15 - 688,529 PPD
 All  Time / Frame : 00:03:14 - 692,221 PPD
 Eff. Time / Frame : 00:03:16 - 684,878 PPD

So on a good one nearly 1m, on a bad one <700K

Anything below 800K ppd on these water-cooled, overclocked 1070s is a bad day. Generally they're chunking along at 1m or thereabouts.

[JohnChodera]

My hope is that since these WUs are short, the variation will average out and we ended up getting the base points about right on average!

We're still working hard to solve the variability problem though---thanks for bearing with us.

~ John Chodera // MSKCC

[astrorob]

just had back-to-back-to-back failures on 13420 on an RX 5500 - the work unit dies right away with "ERROR:Discrepancy: Forces are blowing up! 0 0"

meanwhile 13420 is chugging along on a GTX 1060...

[uyaem]

My hope is that since these WUs are short, the variation will average out and we ended up getting the base points about right on average!

We're still working hard to solve the variability problem though---thanks for bearing with us.

~ John Chodera // MSKCC

I'll try and report here on the forums whenever there are outliers in terms of performance, including hardware used, in case the combination is important (and I think you can't see all those details on the collecting end.
Here's are two more:

13420 (6696, 58, 0) with 420k PPD on GTX 1660S, normal would be 760k in my setup. TPF 4.14 instead of 2.51.
13420 (1808, 39, 0) similar to the above

[gunnarre]

13421 (4308, 53, 0) is making about 10% more PPD than normally expected on my Radeon 7770 HD (TPF 1.36), so I'm guessing this is a normally performing WU that benefits from the boosted Base Credit since July 28th. Edit: After reaching 50% of the WU, PPD is down to about 90k (TPF 1.47) on the 7770 HD which is within what I'd normally expect.

13421 (7733, 52, 0) is making less than half the expected PPD on my RX 580 (TPF 1.13), so this might be a slow one.

[gunnarre]

13421 (5574, 53, 0) Extremely slow on the RX 580, TPF 3.13 (36928 PPD). So the 7770 HD is folding almost three times as fast as the RX 580 at the moment.

[Curt3g]

Here is my history with the 134XX projects, on Windows 10 with a RTX 2060:

Code: Select all

Project		Total run	Total anomalous	Percent anomalous
13412		     3		       0			           0%
13414		     37		      4			           11%
13416		     15		      5			           33%
13418		     9		       0			           0%
13420		     20		      1			           5%

'Anomalous' defined as a step function longer to complete. Nominal is 4 hours, plus or minus a few minutes. The longer ones are all in the 12-13 hour range.

With respect to my machine, the 18 and 20 series appear to be more consistent.

Cheers,

Curt

[pylgrym]

I'm seeing about 30%-40% greater ppd with 13420 than average and the project seems to love my standard GTX-1080, I've seen almost nothing but this project for about the last week.

Pylgrym

[uyaem]

13420 (5569, 98, 1) is also about 50% slower

[themartymonster]

I have had 2 13420 units, first one took over 2 days, this current one says 1 day.
This is using an RTX2070 GPU.
CPU is an AMD 3900x with 64GB Ram.
Work units only run on GPU.
Current unit says 60500 credits for 1 days work.
Previous units were getting 150000 - 250000 for a few hours work, cannot recall which WU they were for comparison.

[bruce]

All projects run either entirely on the CPU threads or (almost) entirely on the GPU. [All GPU projects do require some support from a CPU thread but the heavy computing work is done entriely on the GPU.]

Projects 134xx are GPU projects.

See the 7th column on https://apps.foldingathome.org/psummary The last two digits are either _2n (for GPUs) or _An (for CPUs).